CelebHair - Face/hair segmentations 

Description

Automatic hair (style) analysis has implications in various domains: soft biometrics, fashion, visagisme, avatar generation, just to name a few. However, there are few datasets which provide annotations and images for this task. The purpose of this dataset is to provide segmentation masks (labeled with face, hair and background pixels) for more than 3500 unconstrained, "in-the-wild" face images. The input images are taken from the CelebA . We do not own the input images so you have to contact the authors to obtain permission to use the corresponding input images. 

The segemntation masks correspond to the aligned and cropped png images from the CelebA dataset. Each mask is a bmp file with the same basename as its corresponding input image (for example, the segmentation mask 000567.bmp corresponds to the image 000567.png).

The segmentation masks contain only the following pixels 0 (background pixel), 128 (face area pixel) or 255 (hair area pixel).

Download

Please fill in the agreeement form in order to receive the data!

The database contains the following files:

• Segmentation masks: bmp images corresponding to a subset of the CelebA dataset images.
• Scripts (python): useful scripts to view the segmentation and to select the corresponding input images from the CelebA dataset. 

Terms and conditions

• The database is made public  only for non-commercial purposes.
• This database contains the corresponding segmentation masks (hair, face and skin pixels) for a subset of the images from CelebA dataset. We do not own the corresponding input images for the segmentation masks. You have to contact CelebA authors to obtain permission to also use the corresponding input images 
• You agree not to further copy, publish or distribute any of the segmentation masks included in this database

Citation

If you use these data please cite the following work: 

Borza, Diana, Tudor Ileni, and Adrian Darabant. "A Deep Learning Approach to Hair Segmentation and Color Extraction from Facial Images." International Conference on Advanced Concepts for Intelligent Vision Systems. Springer, Cham, 2018.

Bibtex format:

@inproceedings{borza2018deep,
  title={A Deep Learning Approach to Hair Segmentation and Color Extraction from Facial Images},
  author={Borza, Diana and Ileni, Tudor and Darabant, Adrian},
  booktitle={International Conference on Advanced Concepts for Intelligent Vision Systems},
  pages={438--449},
  year={2018},
  organization={Springer}
}

Related datasets

The input images for the annotation masks are obtained from the CelebA (Large-scale CelebFaces Attributes Dataset).

Other databases with facial hair segmentation or analysis:

• Part Labels Database : the dataset contains 2927 face masks (segmentation into hair/skin/background classes) for images from the LFW (Labeled Faces into the Wild) dataset
• Figaro1k : the database contains 1050 unconstrained images containing various hair textures

Errata

Contact

Please contact dadi (at) cs.ubbcluj.ro Adrian DARABANT or Diana Borza (diana.borza (at) cs.utcluj.ro)  or Tudor Ileni (ileni.tudor (at) gmail.com) for any questions or comments about the database.